Applicability of in silico tools for the prediction of dermal absorption for pesticides

Abstract

Based on the “Human in vitro dermal absorption datasets” published as supporting information to the revised EFSA Guidance on Dermal Absorption, in silico models for prediction of absorption across the skin have been evaluated. For this evaluation, a systematic literature search and review was performed, identifying 288 publications describing mathematical models for prediction of dermal absorption. Eleven models potentially relevant to the regulatory assessment of pesticides and which cover a range of approaches were selected for in depth evaluation. This included three mixture models taking into account physicochemical properties of the co‐formulants such as polar surface area, hydrogen bonding or octanol‐water partition coefficients. Additional data on the pesticidal active substances and information on the composition of some of the formulations covered in the dermal absorption dataset were gathered, as these were required as input parameters for the selected models. The models were implemented with settings reflecting as much as possible realistic exposure scenarios and the experimental conditions under which the measured data were obtained. As the majority of the models predicted either maximum flux or the permeation coefficient, further combination with a model achieving translation into percentage absorption was required. This was done with and without consideration of the lag time. Only one model directly predicted percentage dermal absorption, which was a spreadsheet‐based single substance model taking into consideration several skin parameters, experimental conditions, various physicochemical properties of the active substance and the type of vehicle. Statistical analysis of model predictions revealed overall low concordance with measured values, thereby limiting regulatory acceptance. Additional analysis was performed on the results of two mixture models and the above mentioned complex single substance model which showed moderate correlation between predicted and measured data. Options to improve model performance were discussed and Bayesian random effects modelling was explored to adjust predicted percentage dermal absorption to measured data as was model combination. When taking into account observed uncertainties of predictions, one of the models may provide a Tier 2 tool to estimate dermal absorption value in the absence of adequate experimental data when the predicted values are in the range of 10 to 70%. However, further work is needed to better understand the effect of co‐formulants on dermal absorption of pesticides and to improve model predicitivity.